The Agent Skills Directory

[DYNAMIC_EXECUTION]: The loadFetchers function in scripts/research-loop.js dynamically loads and executes JavaScript files from both the skill's own directory and the user's home directory at ~/.pro-workflow/fetchers/. This mechanism allows for the execution of arbitrary code found in these locations during the skill's initialization.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of data from external sources and compile them into wiki pages that are stored locally, creating a surface for indirect instructions to enter the agent's context.
Ingestion points: Data is fetched from the web (via DuckDuckGo), arXiv, and GitHub through the fetcher scripts in scripts/source-fetchers/.
Boundary markers: None. The scraped content is parsed and directly interpolated into markdown files under the ## Claims and ## Sources headers without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill possesses the ability to write files to the local filesystem (fs.writeFileSync), interact with a database for indexing, and dynamically load external code modules.
Sanitization: Sanitization is limited to HTML tag stripping and whitespace normalization; there is no validation or filtering to remove potential natural language instructions from the sourced text.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch research data from several sources.
Fetches search results and snippets from DuckDuckGo's lite interface.
Downloads academic abstracts and metadata from the official arXiv API (export.arxiv.org).
Retrieves repository information and metadata from the GitHub API (api.github.com).

wiki-research-loop